Can You Predict Who Will Stop Self-Harming in Therapy? A Patient's Emotion Vocabulary in the Attachment Interview Says Yes (r=.46)
- Secondary analysis of an RCT comparing Transference-Focused Psychotherapy (TFP) vs treatment as usual for BPD (N=87 female outpatients, mean age 27-29 years) — assessed emotion word repertoire (EWR) using the electronic Levels of Emotional Awareness Scale (eLEAS) applied to transcripts of the Adult Attachment Interview (AAI)
- Baseline EWR was positively correlated with reduction of non-suicidal self-injury (NSSI) after one year of psychotherapy: r=.46, p<.001 — the strongest single predictor of NSSI reduction reported in this cohort
- The correlation held across both treatment conditions (TFP and TAU), suggesting that emotional vocabulary is a transdiagnostic resource that benefits therapy regardless of the specific approach
- Surprisingly, EWR at baseline was NOT correlated with improvement in other outcome measures (attachment representations, mentalization, personality organization) — the predictive power was specific to behavioral self-harm, not global functioning
Clinicians working with borderline patients know that some patients stop self-harming within months while others persist for years of treatment. The search for baseline predictors — who will respond and who will not — has produced mostly modest results. This study from the Medical University of Vienna offers a novel predictor that aligns with clinical intuition: patients who can articulate their emotions in words at the start of therapy are more likely to stop cutting by the end of the year.
The method is the innovation. The researchers applied a text-based computational scoring system (eLEAS) to transcripts of the Adult Attachment Interview — a well-known research instrument that clinicians rarely use for outcome prediction. The eLEAS counts the diversity, specificity, and granularity of emotion words in open-ended narrative text. It is not a self-report questionnaire — it measures what the patient actually produces in language, not what they claim about their emotional awareness.
How EWR predicts NSSI reduction
The mechanism is clinically coherent. NSSI in BPD functions as an emotion regulation strategy — when internal states become unbearable and unnameable, the body becomes the outlet. A patient with a richer emotion vocabulary has an alternative channel: they can translate somatic distress into symbolic language. This doesn't make the distress disappear, but it shifts the regulation pathway from body-based (cutting, burning) to linguistically mediated (identifying, naming, communicating the state to the therapist).
The r=.46 correlation is substantial for psychotherapy prediction research, where most predictors show r<.20. It means that EWR accounts for roughly 21% of the variance in NSSI reduction — more than diagnosis, symptom severity, or personality organization at baseline.
The specificity of the prediction (NSSI but not other outcomes) makes biological sense. NSSI is the most body-mediated, least linguistically processed behavior in the BPD symptom cluster. Patients with richer emotional vocabulary have less need for the body-based channel precisely because they have an alternative. Other outcomes — mentalization, attachment, personality organization — are more cognitively complex and may require different therapeutic mechanisms to change.
How to use this clinically
You don't need the AAI. The clinical takeaway is not "administer the AAI to every borderline patient" (it requires trained raters and 90-minute interviews). The takeaway is: assess your patient's emotion word repertoire informally in the first sessions. When they describe distressing events, do they use diverse, specific emotion words (abandoned, humiliated, betrayed, contemptuous) or vague, global ones (bad, upset, stressed, overwhelmed)?
For treatment planning: Patients with impoverished EWR may need a longer phase of affect labeling and emotional vocabulary building before moving into interpretive or exposure-based work. The TFP model's emphasis on naming affects in the here-and-now of the therapeutic relationship may be particularly suited to these patients.
For measuring progress: Track whether your patient's spontaneous emotion vocabulary diversifies over treatment. A shift from "I felt bad" to "I felt ashamed and then furious at myself for feeling ashamed" is a measurable signal of therapeutic progress — and may precede behavioral NSSI reduction.
In 87 BPD patients, baseline emotion word repertoire — measured by computational analysis of Adult Attachment Interview transcripts — predicted NSSI reduction after one year of therapy with r=.46. Patients who can name their emotions in words at intake are less likely to need their bodies to regulate them.
Secondary analysis of a single RCT — exploratory, requires replication. Female-only sample (BPD gender distribution is debated, but NSSI patterns differ by gender). The eLEAS scoring system requires AAI transcripts (not routinely collected clinically) and proprietary software. EWR may be confounded with verbal intelligence or education — neither was controlled. The decrease in EWR from baseline to post-treatment is counterintuitive and not fully explained (possibly reflects narrative style changes during therapy rather than genuine vocabulary loss). N=52 at endpoint — substantial attrition from the original 87.